The Efficiency of Hybrid BNN-DWT for Predicting the Construction and Demolition Waste Concrete Strength

Abstract

The current study focuses on two main goals. First, with the use of construction and demolition (C&D) of building materials, a new aggregate was produced and it was utilized for green concrete production. The compressive strength test confirmed the good function of C&DW aggregate concrete. This concrete did not show significant differences with natural sand concrete. Second, Backpropagation neural network (BNN) was adjusted for C&DW concrete strength prediction at different curing times. Although BNN has good accuracy for strength prediction, due to the importance of 28^th day of concrete strength the need to improve the accuracy was felt. So discrete wavelet transform (DWT) was used on BNN and a hybrid network was produced. DWT by filtering the noises can improve the homogeneity of the dataset. The results of DWT-BNN showed that the regression can increase to 98% and the MSE index reduces to 0.001. Continued research has shown that increasing the number of filters to four steps leads to reduced accuracy and increased computational cost. So using DWT-BNN as a hybrid network with one filter can improve prediction ability to the desired level but adding up the number of filters not recommended.